Performance optimization, refactoring and revamping.

- Fixed all test cases to use using(Buffer)
- Fixed all test cases to explicitly specify session

* Refactored DisposableObject

* Added different build directory for TensorflowNET.Examples.GPU

* _FetchHandler: Switched to NPTypeCode

* gfile.cs, Walk(...): Handle case when directory top doesn't exist.

* Tensor.Creation: Perf-opted when creating tensor from NDArray of string

* Graph.cs: refactor and added docs

* Tensor.Creation.cs: perf-ops

* Tensor.Explicit.cs: perf-ops

* Copied globals.regen from NumSharp

- Added supported_numericals_TF_DataType

* Tensor perf-ops and cleanup, Revamped dtypes.cs, some renames.

- Cleanup and docs to all Tensor.cs files
- Changed all uses of System.Convert to NumSharp.Utilities.Converts
- Added all missing types in dtypes.cs
- Renamed tensor.Data<T> to tensor.ToArray<T>, added obsolete message
- Renamed tensor.Data() to tensor.BufferToArray(), added obsolete message
- Made GraphKeys to use const string instead allocating strings at every use of GraphKeys.

* Tensor: Added guards for explicit casts.

* Tensor: Added explicit cast to string

* Tensor.ToArray<T>(): Added support for cases when tensor is scalar.

* Tensor.BufferToArray(): Fixed to use long instead of int.

* TensorShape: Revamped and documented.

* BaseSession: Added Session.run(ITensorOrOperation fetche, params FeedItem[] feed_dict)

* Tensor: renamed _dtype to _override_dtype

- Fixed all locations _dtype is used incorrectly.

* Fixed unit tests

* Tensor.Operations: Reverted commit

* DisposableObject: sorted internal_dispose to properly handle Dispose() calls

* Tensor.DisposeUnmanagedResources: Nullify _handle after delete.

* TensorShape.this[...]: fixed guard check.

* DisposableObject #362

- Fixed all locations _dtype is used incorrectly.